Robotics Reference
In-Depth Information
ing, past experience will often be of great assistance in finding a solution
to a problem.
Schank's early work in this field led to a model for what has be-
come known as Case Based Reasoning—the process of reasoning based
on knowledge of past cases, on precedents. Case Based Reasoning oper-
ates on a knowledge-base of past cases, attempting in the first instance
to find a
source case
that is relevant to a given
target case
.Oneprob-
lem domain ideally suited to this form of reasoning is the law, since laws
are founded very much on the use of precedents, so it is hardly surprising
that Schank's work in the 1980s prompted interest from those researchers
who had taken an interest in how AI might be used to aid legal reasoning.
One early example of a Case Based Reasoning system was Katia
Sycara's PERSUADER program, developed in 1990, which reasoned
about negotiations between a workforce and management. When faced
with a problem in this domain, PERSUADER searched its knowledge-
base of past cases, which included agreements between parties who were
in similar situations to those in the dispute currently under considera-
tion. Those earlier agreements suggested various proposals that might be
successful in the current negotiations, allowing PERSUADER to com-
pare the various past situations with the current dispute (as regards rates
of pay and other features) in order to find the closest match.
Retrieving the most similar case or cases from a program's knowledge-
base is the first of four stages in Case Based Reasoning programs. The
whole cycle of processes is shown in Figure 45.
After retrieving one or more previously experienced cases from its
knowledge-base,
18
a program reuses its knowledge of the relevant previ-
ous case(s) by revising (i.e., amending) the earlier solution(s). Sometimes
an earlier case will be so similar to the current problem that the pro-
gram need make only the smallest of changes in order to be able to apply
the earlier solution and thereby find a solution to the current problem,
while on other occasions more drastic changes will need to be made to
an earlier solution before it is reusable. Finally, whatever parts of the new
case are likely to be useful for future problem solving are retained in the
knowledge-base.
As shown in Figure 45, the program's knowledge-base incorporates
not only knowledge that is specific to past cases, but also some knowl-
edge of a more general nature but still pertaining to the same problem
18
Also called a case-base.
